API Overview

Modules

• api
• api.hypergraph
• api.scripts.wait_process: wait for a process to finish.
• api.transforms
• api.transforms.image
• api.transforms.image.io
• api.transforms.image.photometric
• api.transforms.image.spatial
• api.transforms.random
• api.transforms.semseg
• api.utils
• core.dataset
• core.graph: contains Node and ExecutableGraph.
• core.hypergraph
• core.loss
• core.metric
• core.module
• core.optim
• llutil
• llutil.argparser: parse arguments for functions and command line.
• llutil.board
• llutil.collections
• llutil.config
• llutil.debug
• llutil.dictprocess
• llutil.file_client
• llutil.ignore_me
• llutil.launcher
• llutil.launcher.elastic_multiprocessing
• llutil.launcher.events
• llutil.launcher.launch_agent
• llutil.launcher.launcher
• llutil.launcher.local_elastic_agent
• llutil.logger: logging utilities.
• llutil.multiprocessing: a drop-in replacement for torch.multiprocessing.
• llutil.print
• llutil.pycuda: Integrates PyCUDA to PyTorch and ice.
• llutil.shadow_tb
• llutil.test: helps developers of ice to test.
• llutil.utils

Classes

• dataset.DatasetNode: Automating DataLoader and DataSampler creation and maintainance.
• dataset.ResumableDistributedSampler
• graph.ExecutableGraph: an executable graph.
• graph.GraphOutputCache: a cache for storing and searching forward outputs of nodes.
• graph.InvalidURIError: An Exception raised when valid node URI is expected.
• graph.Node: This class defines the executable node.
• graph.StopAllTasks: An Exception raised to exit current running.
• graph.StopTask: An Exception raised to exit current task.
• hypergraph.Counter: Counter object.
• hypergraph.GlobalCounters: Global counters object.
• hypergraph.HyperGraph: HyperGraph is the container for all nodes.
• hypergraph.Repeat: Repeat a task for a fixed number of times.
• hypergraph.ResumeTaskFailed: raised when task structure does not match during resuming.
• hypergraph.Task: A task is a unit of computation.
• loss.LossMode: An enumeration.
• loss.LossNode
• metric.AverageMeter
• metric.DictMetric
• metric.Meter: value reducer that works recursively.
• metric.MetricNode
• metric.MovingAverageMeter
• metric.SummationMeter
• metric.ValueMeter
• module.ModuleNode: a node that extends torch.nn.Module
• optim.Optimizer: Optimizer configuration API for ice-learn.
• argparser.ArgumentMissingError: Raised when a required argument is missing from command line.
• argparser.ArgumentTypeError: Raised when converting an argument failed.
• argparser.FlexibleArgParser: A flexible and lightweight argument parser that saves loads of code.
• board.BoardWriter
• collections.ConfigDict: stores multi-level configurations easily.
• collections.Counter: count values by group.
• collections.Dict: access dict values as attributes.
• config.Configurable
• debug.SubProcessPdb: Pdb that works from a multiprocessing child
• file_client.BaseStorageBackend: Abstract class of storage backends.
• file_client.CephBackend: Ceph storage backend (for internal use).
• file_client.FileClient: A general file client to access files in different backends.
• file_client.HTTPBackend: HTTP and HTTPS storage bachend.
• file_client.HardDiskBackend: Raw hard disks storage backend.
• file_client.LmdbBackend: Lmdb storage backend.
• file_client.MemcachedBackend: Memcached storage backend.
• file_client.PetrelBackend: Petrel storage backend (for internal use).
• ignore_me.IgnoreMe
• elastic_multiprocessing.MultiprocessContext: PContext holding worker processes invoked as a function.
• elastic_multiprocessing.PContext: The base class that standardizes operations over a set of processes
• elastic_multiprocessing.TailLog: Tails the given log files. The log files do not have to exist when the
• events.Events: Communicate among main process (agent) and subprocesses (workers).
• launch_agent.LaunchConfig: Creates a rendezvous config.
• launcher.ElasticLauncher: A helper Configurable class for torchrun and torch.distributed.launch.
• local_elastic_agent.LocalElasticAgent: An implementation of :py:class:torchelastic.agent.server.ElasticAgent
• pycuda.CUDAModule: Just-In-Time compilation of a set of CUDA kernel functions and device functions from source.

Functions

• api.DatasetNode
• hypergraph.LoadCheckpointTask
• hypergraph.SaveCheckpointTask
• hypergraph.add
• hypergraph.backup_source_files
• hypergraph.get_default_graph
• hypergraph.init_autocast
• hypergraph.init_grad_scaler
• hypergraph.print_forward_output
• hypergraph.run
• hypergraph.set_gradient_accumulate
• io.Load: Args:
• io.imfrombytes: Read an image from bytes.
• io.imread: Read an image.
• io.imwrite: Write image to file.
• io.use_backend: Select a backend for image decoding.
• photometric.Normalize: Normalize the image and convert BGR2RGB.
• photometric.PhotoMetricDistortion: Apply photometric distortion to image sequentially, every dictprocessation
• photometric.ToTensor: Args:
• spatial.Crop: crop a region of interest from the src array.
• spatial.Flip: flip the src image.
• spatial.Pad: Padding a image to target size.
• spatial.Resize: Args:
• spatial.SizeDivisorMultiple: Returns a smallest but larger shape to ensure each edge to be multiple to some number.
• random.RandomChoice
• random.RandomDo
• random.RandomFloats
• random.RandomImage: generate a random image for testing purpose
• random.RandomIntegers
• random.RandomProbabilities
• random.RandomROI: generate random region of interest.
• semseg.LabelToTensor: Convert to tensor (as int64).
• semseg.LoadAnnotation: Load annotations for semantic segmentation.
• semseg.RandomCrop: Crop a pair of images and segmentation labels such that the class is relatively balanced for training.
• utils.parse_devices
• dataset.failsafe_collate: Puts each data field into a tensor with outer dimension batch size
• hypergraph.GlobalCounters.__init__
• hypergraph.LoadCheckpointTask: Load checkpoint from a file.
• hypergraph.SaveCheckpointTask: Save checkpoint to a file.
• argparser.as_dict: helps to regularize input into a dict.
• argparser.as_list: helps to regularize input into list of element.
• argparser.get_hostname
• argparser.is_list_of: Check whether it is a list of some type.
• argparser.is_seq_of: Check whether it is a sequence of some type.
• argparser.is_tuple_of: Check whether it is a tuple of some type.
• argparser.isa: an alias for python built-in isinstance.
• argparser.parse_scalar
• config.clone: clone configurables, containers, and ordinary objects recursively.
• config.configurable: This decorator delays the initialization of cls until freeze().
• config.freeze: freeze configurables recursively.
• config.frozen
• config.has_builder
• config.is_configurable: check if a class or an object is configurable.
• config.make_configurable: This function converts multiple existing classes to configurables.
• config.objattr
• debug.set_trace
• dictprocess.Collect: a predefined DictProcessor that keep only selected entries.
• dictprocess.Compose: a predefined DictProcessor that composes a list of other DictProcessors together.
• dictprocess.dictprocess: a decorator that convert function into a DictProcessor (Callable[[Dict], Dict]).
• file_client.has_method: Check whether the object has a method.
• file_client.mmcv_mkdir_or_exist
• elastic_multiprocessing.start_processes: Starts n copies of entrypoint processes with the provided options.
• elastic_multiprocessing.tail_logfile
• launch_agent.LaunchConfig.__init__
• launch_agent.launch_agent
• launcher.get_current_launcher
• logger.get_logger: set up a simple logger that writes into stderr.
• print.format_size: Format a byte count as a human readable file size.
• print.set_printoptions
• shadow_tb.shadow
• test.requires_n_gpus
• utils.auto_freeze_enabled
• utils.enable_auto_freeze
• utils.in_main_process: Whether current process is worker process or main process.
• utils.init_torch_multiprocessing